label-studio-sdk 1.0.11__tar.gz → 1.0.13__tar.gz

{label_studio_sdk-1.0.11 → label_studio_sdk-1.0.13}/PKG-INFO

@@ -1,6 +1,6 @@
 Metadata-Version: 2.3
 Name: label-studio-sdk
-Version: 1.0.11
+Version: 1.0.13
 Summary: 
 Requires-Python: >=3.9,<4
 Classifier: Intended Audience :: Developers
@@ -28,7 +28,8 @@ Requires-Dist: jsf (>=0.11.2,<0.12.0)
 Requires-Dist: jsonschema (>=4.23.0)
 Requires-Dist: lxml (>=4.2.5)
 Requires-Dist: nltk (>=3.9.1,<4.0.0)
-Requires-Dist: numpy (>=1.26.4,<2.0.0)
+Requires-Dist: numpy (>=1.26.4,<3.0.0)
+Requires-Dist: opencv-python (>=4.9.0,<5.0.0)
 Requires-Dist: pandas (>=0.24.0)
 Requires-Dist: pydantic (>=1.9.2)
 Requires-Dist: pydantic-core (>=2.18.2,<3.0.0)

{label_studio_sdk-1.0.11 → label_studio_sdk-1.0.13}/pyproject.toml

@@ -3,7 +3,7 @@ name = "label-studio-sdk"
 
 [tool.poetry]
 name = "label-studio-sdk"
-version = "1.0.11"
+version = "1.0.13"
 description = ""
 readme = "README.md"
 authors = []
@@ -44,7 +44,8 @@ jsf = "^0.11.2"
 jsonschema = ">=4.23.0"
 lxml = ">=4.2.5"
 nltk = "^3.9.1"
-numpy = ">=1.26.4,<2.0.0"
+numpy = ">=1.26.4,<3.0.0"
+opencv-python = "^4.9.0"
 pandas = ">=0.24.0"
 pydantic = ">= 1.9.2"
 pydantic-core = "^2.18.2"

{label_studio_sdk-1.0.11 → label_studio_sdk-1.0.13}/src/label_studio_sdk/__init__.py

@@ -3,6 +3,7 @@
 from .types import (
     AccessTokenResponse,
     Annotation,
+    AnnotationCompletedBy,
     AnnotationFilterOptions,
     AnnotationLastAction,
     AnnotationsDmField,
@@ -85,6 +86,7 @@ from .types import (
     RedisImportStorageStatus,
     RefinedPromptResponse,
     RefinedPromptResponseRefinementStatus,
+    RotateTokenResponse,
     S3ExportStorage,
     S3ExportStorageStatus,
     S3ImportStorage,
@@ -200,6 +202,7 @@ __all__ = [
     "ActionsCreateRequestSelectedItemsExcluded",
     "ActionsCreateRequestSelectedItemsIncluded",
     "Annotation",
+    "AnnotationCompletedBy",
     "AnnotationFilterOptions",
     "AnnotationLastAction",
     "AnnotationsCreateBulkRequestSelectedItems",
@@ -307,6 +310,7 @@ __all__ = [
     "RedisImportStorageStatus",
     "RefinedPromptResponse",
     "RefinedPromptResponseRefinementStatus",
+    "RotateTokenResponse",
     "S3ExportStorage",
     "S3ExportStorageStatus",
     "S3ImportStorage",

{label_studio_sdk-1.0.11 → label_studio_sdk-1.0.13}/src/label_studio_sdk/converter/converter.py

@@ -11,13 +11,14 @@ from datetime import datetime
 from enum import Enum
 from glob import glob
 from shutil import copy2
-from typing import Optional
+from typing import Optional, List, Tuple
 
 import ijson
 import ujson as json
 from PIL import Image
 from label_studio_sdk.converter import brush
 from label_studio_sdk.converter.audio import convert_to_asr_json_manifest
+from label_studio_sdk.converter.keypoints import process_keypoints_for_coco, build_kp_order, update_categories_for_keypoints, keypoints_in_label_config, get_yolo_categories_for_keypoints
 from label_studio_sdk.converter.exports import csv2
 from label_studio_sdk.converter.utils import (
     parse_config,
@@ -34,6 +35,7 @@ from label_studio_sdk.converter.utils import (
     convert_annotation_to_yolo_obb,
 )
 from label_studio_sdk._extensions.label_studio_tools.core.utils.io import get_local_path
+from label_studio_sdk.converter.exports.yolo import process_and_save_yolo_annotations
 
 logger = logging.getLogger(__name__)
 
@@ -59,6 +61,7 @@ class Format(Enum):
     YOLO_WITH_IMAGES = 14
     COCO_WITH_IMAGES = 15
     YOLO_OBB_WITH_IMAGES = 16
+    BRUSH_TO_COCO = 17 
 
     def __str__(self):
         return self.name
@@ -108,13 +111,13 @@ class Converter(object):
             "description": "Popular machine learning format used by the COCO dataset for object detection and image "
             "segmentation tasks with polygons and rectangles.",
             "link": "https://labelstud.io/guide/export.html#COCO",
-            "tags": ["image segmentation", "object detection"],
+            "tags": ["image segmentation", "object detection", "keypoints"],
         },
         Format.COCO_WITH_IMAGES: {
             "title": "COCO with Images",
             "description": "COCO format with images downloaded.",
             "link": "https://labelstud.io/guide/export.html#COCO",
-            "tags": ["image segmentation", "object detection"],
+            "tags": ["image segmentation", "object detection", "keypoints"],
         },
         Format.VOC: {
             "title": "Pascal VOC XML",
@@ -127,13 +130,13 @@ class Converter(object):
             "description": "Popular TXT format is created for each image file. Each txt file contains annotations for "
             "the corresponding image file, that is object class, object coordinates, height & width.",
             "link": "https://labelstud.io/guide/export.html#YOLO",
-            "tags": ["image segmentation", "object detection"],
+            "tags": ["image segmentation", "object detection", "keypoints"],
         },
         Format.YOLO_WITH_IMAGES: {
             "title": "YOLO with Images",
             "description": "YOLO format with images downloaded.",
             "link": "https://labelstud.io/guide/export.html#YOLO",
-            "tags": ["image segmentation", "object detection"],
+            "tags": ["image segmentation", "object detection", "keypoints"],
         },
         Format.YOLO_OBB: {
             "title": "YOLOv8 OBB",
@@ -167,6 +170,13 @@ class Converter(object):
             "format expected by NVIDIA NeMo models.",
             "link": "https://labelstud.io/guide/export.html#ASR-MANIFEST",
             "tags": ["speech recognition"],
+        },   
+        Format.BRUSH_TO_COCO: {
+
+            "title": "Brush labels to COCO",
+            "description": "Export your brush labels as COCO format for segmentation tasks. Converts RLE encoded masks to COCO polygons.",
+            "link": "https://labelstud.io/guide/export.html#COCO",
+            "tags": ["image segmentation", "brush annotations"],
         },
     }
 
@@ -197,6 +207,7 @@ class Converter(object):
         self._schema = None
         self.access_token = access_token
         self.hostname = hostname
+        self.is_keypoints = None
 
         if isinstance(config, dict):
             self._schema = config
@@ -293,6 +304,19 @@ class Converter(object):
                 upload_dir=self.upload_dir,
                 download_resources=self.download_resources,
             )
+        elif format == Format.BRUSH_TO_COCO:
+            items = (
+                self.iter_from_dir(input_data)
+                if is_dir
+                else self.iter_from_json_file(input_data)
+            )
+            from label_studio_sdk.converter.exports.brush_to_coco import convert_to_coco
+            image_dir = kwargs.get("image_dir")
+            convert_to_coco(
+                items, 
+                output_data, 
+                output_image_dir=image_dir
+            )
 
     def _get_data_keys_and_output_tags(self, output_tags=None):
         data_keys = set()
@@ -355,11 +379,14 @@ class Converter(object):
             and (
                 "RectangleLabels" in output_tag_types
                 or "PolygonLabels" in output_tag_types
+                or "KeyPointLabels" in output_tag_types
             )
             or "Rectangle" in output_tag_types
             and "Labels" in output_tag_types
             or "PolygonLabels" in output_tag_types
             and "Labels" in output_tag_types
+            or "KeyPointLabels" in output_tag_types
+            and "Labels" in output_tag_types
         ):
             all_formats.remove(Format.COCO.name)
             all_formats.remove(Format.COCO_WITH_IMAGES.name)
@@ -376,6 +403,7 @@ class Converter(object):
         ):
             all_formats.remove(Format.BRUSH_TO_NUMPY.name)
             all_formats.remove(Format.BRUSH_TO_PNG.name)
+            all_formats.remove(Format.BRUSH_TO_COCO.name) 
         if not (
             ("Audio" in input_tag_types or "AudioPlus" in input_tag_types)
             and "TextArea" in output_tag_types
@@ -500,6 +528,9 @@ class Converter(object):
                     if "original_height" in r:
                         v["original_height"] = r["original_height"]
                     outputs[r["from_name"]].append(v)
+                    if self.is_keypoints:
+                        v['id'] = r.get('id')
+                        v['parentID'] = r.get('parentID')
 
             data = Converter.get_data(task, outputs, annotation)
             if "agreement" in task:
@@ -616,6 +647,7 @@ class Converter(object):
             os.makedirs(output_image_dir, exist_ok=True)
         images, categories, annotations = [], [], []
         categories, category_name_to_id = self._get_labels()
+        categories, category_name_to_id = update_categories_for_keypoints(categories, category_name_to_id, self._schema)
         data_key = self._data_keys[0]
         item_iterator = (
             self.iter_from_dir(input_data)
@@ -681,9 +713,10 @@ class Converter(object):
                 logger.debug(f'Empty bboxes for {item["output"]}')
                 continue
 
+            keypoint_labels = []
             for label in labels:
                 category_name = None
-                for key in ["rectanglelabels", "polygonlabels", "labels"]:
+                for key in ["rectanglelabels", "polygonlabels", "keypointlabels", "labels"]:
                     if key in label and len(label[key]) > 0:
                         category_name = label[key][0]
                         break
@@ -753,11 +786,22 @@ class Converter(object):
                             "area": get_polygon_area(x, y),
                         }
                     )
+                elif "keypointlabels" in label:
+                    keypoint_labels.append(label)
                 else:
                     raise ValueError("Unknown label type")
 
                 if os.getenv("LABEL_STUDIO_FORCE_ANNOTATOR_EXPORT"):
                     annotations[-1].update({"annotator": get_annotator(item)})
+            if keypoint_labels:
+                kp_order = build_kp_order(self._schema)
+                annotations.append(process_keypoints_for_coco(
+                    keypoint_labels,
+                    kp_order,
+                    annotation_id=len(annotations),
+                    image_id=image_id,
+                    category_name_to_id=category_name_to_id,
+                ))
 
         with io.open(output_file, mode="w", encoding="utf8") as fout:
             json.dump(
@@ -824,7 +868,14 @@ class Converter(object):
         else:
             output_label_dir = os.path.join(output_dir, "labels")
             os.makedirs(output_label_dir, exist_ok=True)
-        categories, category_name_to_id = self._get_labels()
+        is_keypoints = keypoints_in_label_config(self._schema)
+
+        if is_keypoints:
+            # we use this attribute to add id and parentID to annotation data
+            self.is_keypoints = True
+            categories, category_name_to_id = get_yolo_categories_for_keypoints(self._schema)
+        else:
+            categories, category_name_to_id = self._get_labels()
         data_key = self._data_keys[0]
         item_iterator = (
             self.iter_from_dir(input_data)
@@ -901,82 +952,7 @@ class Converter(object):
                         pass
                 continue
 
-            annotations = []
-            for label in labels:
-                category_name = None
-                category_names = []  # considering multi-label
-                for key in ["rectanglelabels", "polygonlabels", "labels"]:
-                    if key in label and len(label[key]) > 0:
-                        # change to save multi-label
-                        for category_name in label[key]:
-                            category_names.append(category_name)
-
-                if len(category_names) == 0:
-                    logger.debug(
-                        "Unknown label type or labels are empty: " + str(label)
-                    )
-                    continue
-
-                for category_name in category_names:
-                    if category_name not in category_name_to_id:
-                        category_id = len(categories)
-                        category_name_to_id[category_name] = category_id
-                        categories.append({"id": category_id, "name": category_name})
-                    category_id = category_name_to_id[category_name]
-
-                    if (
-                        "rectanglelabels" in label
-                        or "rectangle" in label
-                        or "labels" in label
-                    ):
-                        # yolo obb
-                        if is_obb:
-                            obb_annotation = convert_annotation_to_yolo_obb(label)
-                            if obb_annotation is None:
-                                continue
-
-                            top_left, top_right, bottom_right, bottom_left = (
-                                obb_annotation
-                            )
-                            x1, y1 = top_left
-                            x2, y2 = top_right
-                            x3, y3 = bottom_right
-                            x4, y4 = bottom_left
-                            annotations.append(
-                                [category_id, x1, y1, x2, y2, x3, y3, x4, y4]
-                            )
-
-                        # simple yolo
-                        else:
-                            annotation = convert_annotation_to_yolo(label)
-                            if annotation is None:
-                                continue
-
-                            (
-                                x,
-                                y,
-                                w,
-                                h,
-                            ) = annotation
-                            annotations.append([category_id, x, y, w, h])
-
-                    elif "polygonlabels" in label or "polygon" in label:
-                        if not ('points' in label):
-                            continue
-                        points_abs = [(x / 100, y / 100) for x, y in label["points"]]
-                        annotations.append(
-                            [category_id]
-                            + [coord for point in points_abs for coord in point]
-                        )
-                    else:
-                        raise ValueError(f"Unknown label type {label}")
-            with open(label_path, "w") as f:
-                for annotation in annotations:
-                    for idx, l in enumerate(annotation):
-                        if idx == len(annotation) - 1:
-                            f.write(f"{l}\n")
-                        else:
-                            f.write(f"{l} ")
+            categories, category_name_to_id = process_and_save_yolo_annotations(labels, label_path, category_name_to_id, categories, is_obb, is_keypoints, self._schema)
         with open(class_file, "w", encoding="utf8") as f:
             for c in categories:
                 f.write(c["name"] + "\n")

label_studio_sdk-1.0.13/src/label_studio_sdk/converter/exports/brush_to_coco.py

@@ -0,0 +1,332 @@
+"""
+Convert brush annotations and polygon annotations to COCO format.
+This module handles RLE encoded brush masks and converts them to COCO segmentation format.
+"""
+import logging
+import random
+import io
+import os
+import json
+import numpy as np
+import cv2
+from copy import deepcopy
+from datetime import datetime
+
+from label_studio_sdk.converter.utils import ensure_dir, get_annotator
+import label_studio_sdk.converter.brush as brush_module
+
+logger = logging.getLogger(__name__)
+
+
+def generate_random_color():
+    """Generates a random hex color for category visualization"""
+    return '#{:02x}{:02x}{:02x}'.format(random.randint(0, 255), random.randint(0, 255), random.randint(0, 255))
+
+
+def flatten(item):
+    """Flatten nested lists for segmentation format"""
+    if isinstance(item, list):
+        for subitem in item:
+            yield from flatten(subitem)
+    else:
+        yield item
+
+
+def generate_contour_from_rle(rle, original_width, original_height):
+    """Convert RLE encoded masks to COCO segmentation format
+    
+    Args:
+        rle: RLE encoded mask data
+        original_width: Width of the original image
+        original_height: Height of the original image
+        
+    Returns:
+        tuple: (segmentation, bbox_list, area_list) in COCO format
+    """
+    # Ensure rle is bytes, not string
+    if isinstance(rle, str):
+        rle = rle.encode('utf-8')
+        
+    # Decode RLE
+    try:
+        rle_binary = brush_module.decode_rle(rle)
+    except Exception as e:
+        logger.warning(f"Failed to decode RLE: {e}")
+        return [], [], []
+    
+    # Reshape to image dimensions with 4 channels (RGBA)
+    try:
+        reshaped_image = np.reshape(rle_binary, [original_height, original_width, 4])
+    except Exception as e:
+        logger.warning(f"Failed to reshape RLE data: {e}")
+        return [], [], []
+    
+    # Use only the alpha channel for contour detection
+    alpha_channel = np.array(reshaped_image[:, :, 3]).astype('uint8')
+    
+    # Find contours
+    contours, _ = cv2.findContours(
+        alpha_channel, cv2.RETR_EXTERNAL, cv2.CHAIN_APPROX_SIMPLE
+    )
+
+    segmentation = []
+    bbox_list = []
+    area_list = []
+    for contour in contours:
+        # OpenCV contours are in format [[[x1,y1]], [[x2,y2]], ...], 
+        # we need to convert to a flat list [x1,y1,x2,y2,...]
+        contour_points = [point[0] for point in contour]
+        
+        # Flatten to a single list
+        contour_flat = []
+        for point in contour_points:
+            contour_flat.extend(point.tolist())
+        
+        # Check if we have at least 3 points (6 coordinates) for a valid polygon
+        if len(contour_points) >= 3:
+            segmentation.append(contour_flat)
+
+            x, y, w, h = cv2.boundingRect(contour)
+            bbox_list.append([x, y, w, h])   # Format: [top-left x, top-left y, width, height]
+
+            # Compute the area of the contour
+            area = cv2.contourArea(contour)
+            area_list.append(int(area))
+
+    return segmentation, bbox_list, area_list
+
+
+def generate_contour_from_polygon(points, original_width, original_height):
+    """Convert polygon annotations to COCO segmentation format
+    
+    Args:
+        points: List of [x, y] points in percentage format
+        original_width: Width of the original image
+        original_height: Height of the original image
+        
+    Returns:
+        tuple: (segmentation, bbox, area) in COCO format
+    """
+    # Convert from percentage to absolute coordinates
+    modified_points = [
+        [round(x * original_width / 100), round(y * original_height / 100)]
+        for x, y in points
+    ]
+
+    # Initialize lists
+    segmentation = [list(flatten(modified_points))]
+
+    # Convert points to a NumPy array and reshape for OpenCV
+    contour = np.array(modified_points, dtype=np.float32).reshape((-1, 1, 2))
+
+    # Calculate bounding box
+    bbox = cv2.boundingRect(contour)
+    bbox = [int(coordinate) for coordinate in bbox]
+
+    # Compute the area of the contour
+    area = cv2.contourArea(contour)
+
+    return segmentation, bbox, int(area)
+
+
+def convert_to_coco(items, output_dir, output_image_dir=None):
+    """Convert Label Studio annotations to COCO format
+
+    Args:
+        items: Iterable of annotation items
+        output_dir: Directory where to save the resulting COCO JSON
+        output_image_dir: Optional directory for images
+
+    Returns:
+        Path to the generated COCO JSON file
+    """
+    ensure_dir(output_dir)
+    output_file = os.path.join(output_dir, "result_coco.json")
+    
+    if output_image_dir is not None:
+        ensure_dir(output_image_dir)
+    else:
+        output_image_dir = os.path.join(output_dir, "images")
+        os.makedirs(output_image_dir, exist_ok=True)
+
+    # Initialize COCO JSON structure
+    coco_data = {
+        'images': [],
+        'annotations': [],
+        'categories': [],
+        'info': {
+            'year': datetime.now().year,
+            'version': '1.0',
+            'description': 'Converted from Label Studio',
+            'contributor': 'Label Studio Converter',
+            'date_created': str(datetime.now()),
+        }
+    }
+
+    category_map = {}  # To track unique categories
+    annotation_id = 0
+
+    # Process each item
+    for image_id, item in enumerate(items):
+        # Extract image information
+        image_path = None
+        for key, value in item['input'].items():
+            if isinstance(value, str) and any(value.lower().endswith(ext) for ext in ['.jpg', '.jpeg', '.png', '.gif', '.bmp']):
+                image_path = value
+                break
+        
+        if not image_path:
+            logger.warning(f"No image path found for item {item.get('id')}. Skipping.")
+            continue
+
+        image_info = {
+            'id': image_id,
+            'file_name': os.path.basename(image_path),
+            'path': image_path,
+            'width': None,  # Will be filled from annotations
+            'height': None  # Will be filled from annotations
+        }
+
+        # Skip tasks without annotations
+        if not item['output']:
+            logger.warning(f"No annotations found for item {item.get('id')}")
+            continue
+
+        valid_annotations = False
+
+        # Process all annotations
+        for output_key, annotations in item['output'].items():
+            for annotation in annotations:
+                # Get image dimensions from the first annotation
+                if image_info['width'] is None and 'original_width' in annotation:
+                    image_info['width'] = annotation['original_width']
+                    image_info['height'] = annotation['original_height']
+                
+                # Identify annotation type and extract labels
+                annotation_type = annotation.get('type', '').lower()
+                labels = []
+                
+                if 'brushlabels' in annotation:
+                    labels = annotation['brushlabels']
+                    type_key = 'brushlabels'
+                elif 'polygonlabels' in annotation:
+                    labels = annotation['polygonlabels']
+                    type_key = 'polygonlabels'
+                elif 'rectanglelabels' in annotation:
+                    labels = annotation['rectanglelabels']
+                    type_key = 'rectanglelabels'
+                elif 'labels' in annotation:
+                    labels = annotation['labels']
+                    type_key = 'labels'
+                else:
+                    logger.debug(f"Unsupported annotation type: {annotation_type}")
+                    continue
+                
+                # Process each label
+                for label in labels:
+                    # Add new categories to the map
+                    if label not in category_map:
+                        category_id = len(category_map) + 1
+                        category_map[label] = category_id
+                        coco_data['categories'].append({
+                            'id': category_id,
+                            'name': label,
+                            'supercategory': '',
+                            'color': generate_random_color(),
+                            'metadata': {}
+                        })
+                    
+                    category_id = category_map[label]
+                    
+                    # Convert annotation based on type
+                    if 'rle' in annotation and type_key == 'brushlabels':
+
+                        # check required keys exist
+                        if not all(k in annotation for k in ['rle', 'original_width', 'original_height']):
+                            logger.warning(f"Missing required keys for RLE annotation. Skipping.")
+                            continue
+
+                        # Process brush annotation (RLE encoded mask)
+                        segmentations, bboxes, areas = generate_contour_from_rle(
+                            annotation['rle'], 
+                            annotation['original_width'], 
+                            annotation['original_height']
+                        )
+                        
+                        # Create separate annotation for each contour
+                        for i, (segmentation, bbox, area) in enumerate(zip(segmentations, bboxes, areas)):
+                            coco_annotation = {
+                                'id': annotation_id,
+                                'image_id': image_id,
+                                'category_id': category_id,
+                                'segmentation': [segmentation],  # COCO format requires double nesting
+                                'area': area,
+                                'bbox': bbox,
+                                'iscrowd': 0,
+                                'annotator': get_annotator(item, "unknown")
+                            }
+                            coco_data['annotations'].append(coco_annotation)
+                            annotation_id += 1
+                            valid_annotations = True
+                    
+                    elif 'points' in annotation and type_key == 'polygonlabels':
+                        # check required keys exist
+                        if not all(k in annotation for k in ['points', 'original_width', 'original_height']):
+                            logger.warning(f"Missing required keys for polygon annotation. Skipping.")
+                            continue
+                        # Process polygon annotation
+                        segmentation, bbox, area = generate_contour_from_polygon(
+                            annotation['points'], 
+                            annotation['original_width'], 
+                            annotation['original_height']
+                        )
+                        
+                        coco_annotation = {
+                            'id': annotation_id,
+                            'image_id': image_id,
+                            'category_id': category_id,
+                            'segmentation': segmentation,
+                            'area': area,
+                            'bbox': bbox,
+                            'iscrowd': 0,
+                            'annotator': get_annotator(item, "unknown")
+                        }
+                        coco_data['annotations'].append(coco_annotation)
+                        annotation_id += 1
+                        valid_annotations = True
+                    
+                    elif annotation_type == 'rectanglelabels' or type_key == 'labels':
+                        # check required keys exist
+                        if not all(k in annotation for k in ['x', 'y', 'width', 'height', 'original_width', 'original_height']):
+                            logger.warning(f"Missing required keys for rectangle annotation. Skipping.")
+                            continue
+
+                        # Convert from percentage to absolute coordinates
+                        x = annotation['x'] * annotation['original_width'] / 100
+                        y = annotation['y'] * annotation['original_height'] / 100
+                        w = annotation['width'] * annotation['original_width'] / 100
+                        h = annotation['height'] * annotation['original_height'] / 100
+                        
+                        coco_annotation = {
+                            'id': annotation_id,
+                            'image_id': image_id,
+                            'category_id': category_id,
+                            'segmentation': [],  # Empty for bounding boxes
+                            'area': w * h,
+                            'bbox': [x, y, w, h],
+                            'iscrowd': 0,
+                            'annotator': get_annotator(item, "unknown")
+                        }
+                        coco_data['annotations'].append(coco_annotation)
+                        annotation_id += 1
+                        valid_annotations = True
+        
+        # Add image to the dataset if it has valid dimensions
+        if image_info['width'] is not None and image_info['height'] is not None and valid_annotations:
+            coco_data['images'].append(image_info)
+
+    # Write COCO JSON to file
+    with io.open(output_file, 'w', encoding='utf8') as f:
+        json.dump(coco_data, f, indent=2, ensure_ascii=False)
+    
+    return output_file